Topology Optimization of Electromagnetic 2D Structures for Antennas

The growing demand for high-performance antennas requires a fast and efficient development of radio frequency components. Manually pushing the performance to the limit is time-consuming, and conventional optimization approaches with parameter sizing are rigid and need more flexibility. The study investigates the potential of topology optimization (TO) in radio frequency component development - a technique primarily applied in mechanics. Four different optimization algorithms are compared. They are a genetic algorithm, particle swarm optimization, a surrogate model-based optimizer working with deep neural networks, and a newly developed population-based optimizer that leverages electromagnetic field strengths. Additionally, two TO approaches are tested - i.e. a fragment/pixel-based method and a normalized Gaussian network. The algorithms and approaches are compared to each other and to a benchmark structure. The test object used for the investigations is a frequency-selective surface. The genetic algorithm combined with the pixel-based approach demonstrated the fastest convergence and best performance.